Data to 20 November 2026
About

Pre-registered research

SIP studies: pre-registered specification

Written 2026-10-02 by lane 2, before any SIP result was computed. The engine, costs and tax rules are those of the portfolio lab v2 (portfolio_lab_v2_spec.md, sections 3, 7 and 8 and Appendix B) and the library (portfolio_library.md). Changes after this date go under “Amendments” with the date and the reason.

1. Questions

  1. For each portfolio in the library, what range of after-tax returns did a monthly SIP deliver over 3, 5, 7, 10, 15 and 20 years, depending on when it started?
  2. How often did a SIP do worse than the same instalments put in bank fixed deposits, and worse than inflation?
  3. Do the popular SIP refinements (step-ups, a lucky date of the month, valuation-aware amounts, STPs) change the outcome, and what does stopping a SIP in a crash cost?
  4. Can I reproduce published Indian SIP studies, and where I differ, why?

2. Pre-registered predictions (graded in the results log)

3. The plain SIP

4. Windows and overlap

5. What is reported for each portfolio and horizon

6. Variants

Run on a headline set of 12 portfolios: n50, n500, mid150, small250, sixty_forty, sixty_twenty_twenty, equal_thirds, permanent, all_weather, gtaa, dual_momentum, factor_four.

VariantRule
SIP against lump sumFor each start month and horizon, the SIP’s XIRR against the CAGR of investing the same total at the start. This is the comparison people usually make, but the two are not alternatives for the same investor; the STP study below is the fair one.
Step-up SIPInstalment grows 5% (and 10%) each year on the anniversary of the first instalment.
Date of monthFirst session on or after the 1st, 5th, 10th, 15th, 20th, 25th and 28th. Paired against the 1st.
Dynamic SIP (valuation-aware)Cash-neutral: the investor sets aside Rs 10,000 a month. The amount invested is 1.5× when the Sensex CAPE10 (IIMA, previous month-end) is at or below its expanding 33rd percentile, 1× in the middle, and 0.5× at or above the 67th percentile. Money not invested waits in the liquid sleeve (taxed as a liquid fund), which funds the 1.5× months; if it runs short, only what is there is invested. Everything, including the buffer, counts at the end. A second version uses the Nifty 500 earnings yield minus the 10-year G-sec yield. Applied to n50 and n500 only.
STP from liquid to equityA lump sum of Rs 12 lakh starts in the liquid sleeve and moves to the equity portfolio in 6 (and 12) equal monthly transfers, against investing it all on day one. Compared at 1, 3, 5 and 10 years. Equity portfolios in the headline set only.
Pausing in drawdowns (the behaviour gap)The SIP stops when the Nifty 500 is more than 20% below its previous peak at a month-end and restarts when the drawdown is back above −10%. Two behaviours: (a) the skipped instalments stay in the liquid sleeve, never invested in equity; (b) the backlog is invested in one go on restart. Both against the uninterrupted SIP.
Exemption offSection 3’s exemption switched off.

7. Replication checks

The engine is run in a “replication mode” (no costs, no tax, the published index) and compared with these published studies:

StudyWhat they reportMy closest reproductionExpected differences
Freefincal (Pattabiraman 2020): Nifty 50 TRI, Jul 1999 to Jan 202010-year rolling SIP XIRR fell from above 20% (2009 endings) to about 10% (2020 endings)Nifty 50 TRI, same dates, 10-year rolling XIRR by end monthinstalment date conventions
WhiteOak Capital (2024): Sensex TRI, Sep 1996 to May 2024minimum SIP XIRR 4.6% at 10 years, 7.4% at 15; 79% / 90% of windows above 12% at 10 / 15 yearsNifty 50 TRI from Jul 1999 (I do not hold the Sensex TRI)different index and a shorter sample: I expect the same shape, not the same numbers
Geojit (2024): Sensex, Sep 1996 to Jul 202410-year median 14.3%, minimum 4.1%; 15-year median 14.3%, minimum 7.0%as aboveas above, and Geojit does not say whether it used the TRI

A reproduction counts as a match when the same index and dates give a median within 0.5 points and a minimum within 1 point. Where the index differs, I report both and explain the gap; I do not tune anything to close it. A study whose numbers I cannot reproduce on the same index is reported as such.

8. Outputs

Derived tables .cache/derived/sip_*.parquet and bundles under lab/sip/*, documented in FRONTEND_HANDOFF.md (lane 2 section): the outcome distributions by portfolio and horizon, the per-start-month XIRR series for the explorer, and one table per variant.

9. Known limits

Amendments

2026-10-07 (owner’s request, after results): showcase portfolios and two descriptive views. These add description only; no prediction (S1-S5) is regraded and nothing in sections 3-6 changes.

2026-10-08 (owner’s request): the second set of studies, written before any of them was computed. Code: lab/sip_more.py; bundles lab/sip/pain, lab/sip/timing, lab/sip/amount, lab/sip/years. Rs 10,000 a month on the first session of each month unless stated; common starts from 2005-04.

Definitions.

Predictions.

2026-10-08 (owner’s request): study A, the long history from 1995, measured data only. Written before any of it was computed. Code: lab/sip_long.py; bundle lab/sip/long; page /lab/sip/long/; methods note docs/methods/long_history.md.

Series.

Portfolios. Nifty 500, Nifty 50, gold alone, the synthetic G-sec alone, 60/40 (Nifty 500 / G-sec), 60/20/20 and equal thirds (with gold), rebalanced each January. Plain SIP, after tax, horizons 3 to 20 years; the to-date paths of the pain studies as well.

Validation, with pass marks set now. (1) Gold join: correlation of monthly log changes between RBI’s series and the lab’s from 2005 at least 0.95 (it was 0.96 over 2000-2025 before any study result; input check). (2) Synthetic G-sec against the NSE 5-year G-sec index, 2001 on: correlation of monthly returns at least 0.8, and the synthetic’s higher volatility consistent with its longer duration. Both are reported on the page whatever they show.

Predictions.

2026-10-08 (owner’s request): the FD rate as section 5 names it, written before the rerun. Section 5 named banks’ deposit rate for terms over one year as the FD rate, with the 364-day T-bill “until lane 1 adds it to the warehouse”. It was never added, and the verification sweep of 2026-10-08 measured the proxy’s gap: banks’ 1-3 year rate (top of RBI’s range) averaged 1.33 points above it over FY1993-FY2010 (4.17 in 1997-98), and RBI’s weekly rate over one year (maximum) 0.76 points above it over 2012-2026. The FD rate is now (lab/deposit_rates.py):

The 10-year G-sec is NSE’s index from 2001 (owner decision, 2026-10-09, after results). The lab’s gsec10, which these studies use, is now NSE’s Nifty 10 yr Benchmark G-Sec (total return) from its base date, 2 January 2001: the niftyindices history the Data bank holds, and Kite’s copy of the same index for 2001-2010 and for days missing from ours (Kite matches NSE on 3,840 shared days; a bad stretch in November 2015 and one day in June 2024 take NSE’s values, and one Kite-only spike, 12 March 2009, is rebuilt from the days either side). Before 2001 the bond rebuilt from yields stays, scaled to meet NSE’s first value. Rules and checks: clean_gs10 and gsec10_series in pipeline/tipsheet/lab/universe.py; Kite’s raw series is kept unedited in pipeline/tipsheet/reference/kite_nifty_gs_10yr.csv. Why: the rebuilt bond is a theoretical one: always exactly ten years to run, priced at the curve’s 10-year yield. NSE’s index holds the actual benchmark bond, which a gilt fund can buy. The rebuilt bond ran 0.8 points a year above it over 2001-2010 and 1.6 points over 2011-2025 (monthly correlation 0.95 and 0.75). Its daily tail, FBIL’s par yield through India Data Hub, also posts a session or two after NSE. All three studies are rerun. S14 now reads “the G-sec SIP” for “the synthetic G-sec SIP” and is regraded; every earlier grade stays in the log. Mixes holding the G-sec still start in June 1996, on the rebuilt bond until 2001.

2026-10-09 (owner’s request): study B, the reference view, written before any of it was computed. The owner asked for the SIP pages to be rebuilt as a reference: every holding period from 1 to 20 years, comparisons with G-sec and gold as well as FDs, real returns, and whether SIP returns have fallen with the decade the SIP started in, back to the Sensex’s first year. Code: lab/sip_ref.py; bundle lab/sip/ref.

Every holding period. The plain SIP of section 3, after tax, at every whole horizon from 1 to 20 years, for (a) the long-history portfolios of study A from their first start month and (b) the showcase set from 2005-04. Nothing else in sections 3 to 5 changes. Start months are those that leave the full horizon, so the 1-year rows have the most windows and the 20-year rows the fewest.

The Sensex from 1979, index level.

Predictions.

Results log

2026-10-09: study B, the reference view (S17-S22; data to 2026-10-07, Sensex to 2026-10-01)

Inputs. The Sensex total-return rebuild passed its pass mark: monthly correlation with the Nifty 50 TRI from 2001 of 0.993 (mark 0.97) and a CAGR gap of 0.11 points (mark 1). The CPI link at October 1988 joins the OECD series to the site’s at a ratio of 0.657; over 426 overlapping months their year-on-year rates correlate at 0.963.

Predictions.

Other results. At fifteen years the price-only Sensex medians fall every decade (16.9%, 14.2%, 12.2%, 11.7%; the last on 22 start months). A Sensex SIP started in February 1992 spent 125 months in a row behind the same money in FDs on the price index (59 months with dividends). 97% of Sensex SIPs since 1979 were under water at some month-end; the deepest, a SIP started in October 2007, was 48% below the money put in on 27 October 2008.

2026-10-09: rerun on NSE’s 10-year G-sec (all three studies; data to 2026-10-06)

2026-10-08: rerun with the FD rate from RBI deposit rates (all three studies)

The FD rate is now banks’ deposit rate (amendment above), with no stretch on the T-bill proxy (meta.fd.proxy_used_from is empty; the weekly series was fetched from India Data Hub for this local run, as the warehouse will hold it). Median after-tax FD XIRR of a 10-year SIP: 5.2% (was 4.6%). Changes in the FD comparisons; nothing else moves:

2026-10-08: study A, from 1995 (S13-S16; data to 2026-10-06)

lab/sip_long.py, 178 s: 10,260 windows and 2,536 to-date paths over seven portfolios. Validation passed both pre-set marks: the gold join’s monthly changes correlated 0.964 with the lab’s gold from June 2005 (mark 0.95); the synthetic 10-year G-sec’s monthly returns correlated 0.834 with NSE’s 5-year G-sec index from October 2001 (mark 0.80), 7.98% a year against 7.80%, volatility 6.8% against 3.8%.

2026-10-08: the second set (S6-S12; data to 2026-10-06)

lab/sip_more.py, 199 s on 14 cores: 3,855 to-date paths (15 portfolios x every start month to 2026-09), 8,415 dip-rule windows, 2,805 weekly windows, 10,008 step-up windows at 15%, 1,668 regular-plan windows. Bundles lab/sip/pain, timing, amount, years; pages under /lab/sip/.

2026-10-02: first run (starts April 2005 to September 2023; data to 2026-09-30)

136 portfolios × up to 222 start months × six horizons: 112,608 SIP windows, plus the variants and the longest-history view. Code: pipeline/tipsheet/lab/sip.py; tables .cache/derived/sip_*; bundles lab/sip/*. All XIRRs are after costs and tax, on full redemption at the end of the window, with the equity exemption on. The FD hurdle is the 364-day T-bill proxy (deposit-rate data not yet in the warehouse), after 30% tax and cess each year: its median after-tax yield was 4.5% to 4.9% depending on the horizon. “Independent periods” is the number of non-overlapping windows that fit in the sample; percentiles describe history and are not probabilities.

The plain SIP (Rs 10,000 a month), common starts from April 2005.

PortfolioHorizonIndependent periods5th pctMedianWorst (start)Below FDBelow inflation
Nifty 5005 years43.913.5−6.1 (2015-04)8.1%20.2%
Nifty 50010 years29.112.83.4 (2010-04)0.7%0.7%
Nifty 50015 years18.912.96.1 (2005-04)0%1.3%
Nifty 5010 years28.512.03.4 (2010-04)0.7%1.4%
Smallcap 2505 years4−2.916.7−17.8 (2015-04)23.2%28.3%
Smallcap 25010 years26.714.5−1.1 (2010-04)2.9%3.6%
60/405 years45.511.20.6 (2015-04)3.5%19.7%
60/4010 years29.010.95.9 (2010-04)0%0%
60/20/205 years47.411.41.1 (2015-04)1.5%11.1%
Equal thirds5 years46.610.54.5 (2015-04)0.5%7.6%
Permanent5 years46.69.34.7 (2015-04)0%7.1%
GTAA10 years25.77.85.4 (2009-08)0%0%
100% G-sec5 years45.06.44.7 (2018-11)0%44.9%

Variants (headline set of 12 portfolios).

Replication (Nifty 50 TRI, no costs, no tax, instalment on the first session of each month from July 1999).

HorizonWindowsMinMedianShare above 12%
10 years2073.9%13.7%73%
15 years1476.9%13.4%86%
20 years8710.9%13.6%91%

Predictions.

2026-10-03: audit corrections (register rows L21–L24, V3, V7, D4); rules unchanged

2026-10-07, wording only: the site is run by one person, so “we” and “our” now read “I” and “my”. No definition, test, threshold or result changed.

This is docs/research/sip_studies_spec.md. The specification was committed before any result was computed; changes after that are logged in it with dates and reasons.